Remote Sensing of Environment - misclab.umeoce.maine.edumisclab.umeoce.maine.edu/documents/Siegeletal2003.pdf · from satellite ocean color sensors that sample the entire Earth surface

Remote Sensing of Environment 135 (2013) 77–91

Contents lists available at SciVerse ScienceDirect

Remote Sensing of Environment

j ourna l homepage: www.e lsev ie r .com/ locate / rse

Regional to global assessments of phytoplankton dynamics from the SeaWiFS mission

D.A. Siegel a,⁎, M.J. Behrenfeld b, S. Maritorena a, C.R. McClain c, D. Antoine d, S.W. Bailey c, P.S. Bontempi e,E.S. Boss f, H.M. Dierssen g, S.C. Doney h, R.E. Eplee Jr. c, R.H. Evans i, G.C. Feldman c, E. Fields a, B.A. Franz c,N.A. Kuring c, C. Mengelt j, N.B. Nelson a, F.S. Patt c, W.D. Robinson c, J.L. Sarmiento k, C.M. Swan a,1,P.J. Werdell c, T.K. Westberry b, J.G. Wilding c, J.A. Yoder h

a University of California, Santa Barbara, Santa Barbara, CA 93106-3060, USAb Oregon State University, Corvallis, OR, 97331-2902, USAc NASA Goddard Space Flight Center, Greenbelt, MD 20771, USAd Laboratoire d'Océanographie de Villefranche, 06238 Villefranche sur Mer Cedex, Francee NASA Headquarters, Washington, DC 20546, USAf University of Maine, Orono, ME 04469-5706, USAg University of Connecticut, Groton, CT 06340-6048, USAh Woods Hole Oceanographic Institution, Woods Hole, MA 02543, USAi University of Miami, Miami, FL 33149-1098, USAj Ocean Studies Board, The National Academies, Washington, DC 20001, USAk Princeton University, Princeton, NJ 08540-6654, USA

⁎ Corresponding author. Tel.: +1 805 893 4547.E-mail address: [email protected] (D.A. Siegel).

1 Present address: Institute of Biogeochemistry and PoSwitzerland.

0034-4257/$ – see front matter © 2013 Elsevier Inc. Allhttp://dx.doi.org/10.1016/j.rse.2013.03.025

a b s t r a c t
a r t i c l e i n f o
Article history:Received 18 January 2012Received in revised form 13 March 2013Accepted 25 March 2013Available online xxxx

Keywords:Ocean colorSeaWiFSPhytoplanktonColored dissolved organic matterDecadal trends

Photosynthetic production of organic matter by microscopic oceanic phytoplankton fuels ocean ecosystemsand contributes roughly half of the Earth's net primary production. For 13 years, the Sea-viewing WideField-of-view Sensor (SeaWiFS) mission provided the first consistent, synoptic observations of global oceanecosystems. Changes in the surface chlorophyll concentration, the primary biological property retrievedfrom SeaWiFS, have traditionally been used as a metric for phytoplankton abundance and its distributionlargely reflects patterns in vertical nutrient transport. On regional to global scales, chlorophyll concentrations co-varywith sea surface temperature (SST) because SST changes reflect light and nutrient conditions. However, theoceanmay be too complex to bewell characterized using a single index such as the chlorophyll concentration. Asemi-analytical bio-optical algorithm is used to help interpret regional to global SeaWiFS chlorophyll observa-tions from using three independent, well-validated ocean color data products; the chlorophyll a concentration,absorption by CDM and particulate backscattering. First, we show that observed long-term, global-scale trendsin standard chlorophyll retrievals are likely compromised by coincident changes in CDM. Second, we partitionthe chlorophyll signal into a component due to phytoplankton biomass changes and a component caused byphysiological adjustments in intracellular chlorophyll concentrations to changes in mixed layer light levels.We show that biomass changes dominate chlorophyll signals for the high latitude seas and where persistentvertical upwelling is known to occur, while physiological processes dominate chlorophyll variability overmuch of the tropical and subtropical oceans. The SeaWiFS data set demonstrates complexity in the interpretationof changes in regional to global phytoplankton distributions and illustrates limitations for the assessment ofphytoplankton dynamics using chlorophyll retrievals alone.

© 2013 Elsevier Inc. All rights reserved.

1. Introduction

The open ocean accounts for nearly 70% of the Earth's surface andphytoplankton dominate the autotrophic production of the marineecosystems, accounting for roughly half of the Earth's annual net pri-mary production (e.g., Behrenfeld et al., 2001; Field et al., 1998). This

llutant Dynamics, ETH Zürich,

rights reserved.

conversion of CO2 into organic matter fuels the metabolic demands ofmarine ecosystems and drives ocean biogeochemical cycles. Global phy-toplankton production depends on the availability of sunlight and nutri-ents (e.g., nitrogen, phosphorous, iron), and thus is sensitive to changesin the physical processes regulating these resources (e.g., Behrenfeld etal., 2001, 2006; Boyce et al., 2010; Martinez et al., 2009; Vantrepotte &Mélin, 2009; Yoder & Kennelly, 2003).

The response of phytoplankton to environmental changes can berapid. Under light- and nutrient-replete growth conditions, phyto-plankton abundances can double within a single day, leading tointense blooms. These rapid increases in phytoplankton standing

http://dx.doi.org/10.1016/j.rse.2013.03.025

mailto:[email protected]


http://www.sciencedirect.com/science/journal/00344257

Wavelength (nm)

Con

trib

utio

n to

Spe

ctra

l Abs

orpt

ion

1.0

0.8

0.6

0.4

0.2

0.0400 500 600 700

Detritus CDOM Phytoplankton Water

Fig. 1. Global mean fractional contributions made by phytoplankton (aph(λ); red),colored dissolved organic matter (CDOM; ag(λ); blue), detrital particulates(adet(λ); blue), and seawater (aw(λ); light blue) to the total absorption(atot(λ) = adet(λ) + aph(λ) + ag(λ) + aw(λ)) at that same wavelength. The gray enve-lopes are 95% confidence intervals for themean spectra. A total of 371 coincident observa-tions of adet(λ), aph(λ) and ag(λ) from theupper 30 mare used to develop the global, openocean composite of fractional contributions to spectral light absorption. Absorption spec-tra are compiled from the Atlantic, Pacific, Indian and Southern Oceans (CLIVAR cruisesA20, A22, P16N, P16S, P18 and I8S, the African Monsoon Multidisciplinary Analyses(AMMA) cruise to the equatorial Atlantic and the Bermuda Atlantic Time Series Study(BATS)). Methods and further details are presented in Nelson et al. (1998), Nelson et al.(2007), Nelson et al. (2010), Swan et al. (2009) and Nelson and Siegel (2013). (For inter-pretation of the references to color in this figure legend, the reader is referred to the webversion of this article.)

78 D.A. Siegel et al. / Remote Sensing of Environment 135 (2013) 77–91

stocks are often followed by rapid declines, as resources are exhaustedand grazers and other export processes consume phytoplankton.This balance between phytoplankton growth and loss processes controlsthemagnitude of the response of phytoplankton populations to environ-mental changes on seasonal to decadal time scales (e.g., Behrenfeld,2010; Behrenfeld et al., 2006, 2008; Falkowski et al., 1998; Martinez etal., 2009; Polovina et al., 2008; Siegel et al., 2002a).

Observing changes in global phytoplankton distributions is chal-lenging at best. Research vessels travel at roughly the speed of a bicycle,have limited geographic range, and are costly to operate, making themimpractical for long-term, sustained, global surveying. Autonomoussampling platforms equipped with sensors capable of measuring rele-vant ocean ecosystem parameters have been developed (e.g., Bishopet al., 2002; Johnson et al., 2009) but these advances are relatively re-cent and appropriate data over the global ocean are not yet available.Global observations of phytoplankton dynamics have resulted largelyfrom satellite ocean color sensors that sample the entire Earth surfacein just a few days (McClain, 2009).

The longest and most complete satellite ocean color record stemsfrom the Sea viewing Wide-Field of view Sensor (SeaWiFS), whichwas launched on August 1, 1997 and operated until December 14,2010. A key objective for the SeaWiFS mission was the determinationof surface layer phytoplankton chlorophyll a concentration (hereafterreferred to simply as chlorophyll). The standard SeaWiFS chlorophyllproduct is based on an empirical band-ratio algorithm (O'Reilly et al.,1998). The basic principle of the algorithm is that ocean remote sensingreflectance at blue wavelengths decreases relative to the green wave-lengths, as phytoplankton concentrations increase, due to absorptionby chlorophyll and other constituents. The ratio between these twobands is used to quantify the chlorophyll concentration.

An underlying assumption in the band ratio approach is thatchanges in ocean color can be described using a single index — thechlorophyll concentration (Morel, 1988; Smith & Baker, 1978). Thispresumes that all optically active materials covary with the chlorophyllconcentration, which may not hold under all situations or for futureoceans (Dierssen, 2010; Lee et al., 2002, 2010; Morel et al., 2010; Saueret al., 2012; Siegel et al., 2002b, 2005a; Szeto et al., 2011). Non-livingma-terials, such as Colored Dissolved Organic Matter (CDOM) and nonviable,detrital particulate materials, dominate ocean color signals for somewavelengths and these materials can vary independently from changesin chlorophyll concentration (Bricaud et al., 2011; Morel et al., 2010;Nelson & Siegel, 2013; Siegel et al., 2005a, 2005b). An example of the con-tributions made by different optical components to the mean total ab-sorption spectrum is shown in Fig. 1 where the component absorptionspectrum observations are from a consistent, open ocean, in situ dataset that spans the upper layers (≤30 m) from all of the major ocean ba-sins (N = 371; data from Nelson et al., 1998, 2007, 2010; Swan et al.,2009; Nelson & Siegel, 2013). The gray envelopes in Fig. 1 represent the95% confidence interval for each mean estimate. For wavelengths lessthan 490 nm, CDOM is by far the most important factor contributing tototal absorption in the open ocean and, on the average, 60% of the totallight absorption at 400 nm is due to CDOM (Fig. 1). Phytoplankton ab-sorption never dominates the mean spectral light absorption budgetand only for wavelengths greater than 490 nm does its contribution ap-proach that of CDOM. Throughout the visible spectrum, the contributionof CDOM is also much greater than the absorption by inanimate, detritalparticulates,whichmake amuch smaller contribution throughout the vis-ible spectrum (detrital absorption is b10% of the total absorptioncoefficient). Thus, nonliving optical signals are an important considerationin the assessment of phytoplankton dynamics from global ocean colorobservations.

Phytoplankton optical properties are also functions of their size,shape, cellular pigment composition and concentration, which are inturn regulated by the assemblage of phytoplankton species present andtheir physiological state (e.g., Behrenfeld et al., 2005, 2008; Bricaud etal., 2004; Ciotti et al., 2002; Dierssen, 2010; Falkowski, 1984; Kirk, 1994;

Morel & Bricaud, 1981; Stramski et al., 2004). In particular, phytoplanktoncan adjust their intracellular chlorophyll content in response to light andnutrients (e.g., Behrenfeld et al., 2005; Laws & Bannister, 1980). Whengrowth irradiance decreases andnutrients remain replete, phytoplanktonincrease their chlorophyll content tomore efficiently capture light.Whennutrients become scarce, they reduce their chlorophyll in direct propor-tion to their nutrient-defined growth rate (e.g., Halsey et al., 2010; Laws& Bannister, 1980). Together these physiological adjustments can intro-duce more than a ten-fold variation in phytoplankton chlorophyll to car-bon ratios, which will hinder our ability to interpret changes inphytoplankton biomass from observations of chlorophyll concentrations.

Here we present a synthesis of SeaWiFS observations focused ondocumenting and interpreting regional- to global-scale change in oceanphytoplankton distributions. We summarize how the SeaWiFS missionachieved a climate-quality record of ocean remote sensing reflectancespectra (referred to as ‘ocean color’) and how ocean color observationsare used to retrieve phytoplankton chlorophyll concentrations. Wediscuss changes in the ocean biosphere indicated by the standardSeaWiFS chlorophyll algorithm product and show that spurious inter-pretationsmay result fromnot assessing other important bio-optical sig-nals. Our analyses highlight the spatial heterogeneity of phytoplanktonresponses to ocean climate variations, demonstrate the importance ofnon-phytoplankton absorption in ocean color signals and illustrate therelative importance of physiology and biomass changes on observedchlorophyll concentrations and the interpretation of phytoplanktondynamics from ocean color observations. We conclude with a discussionof steps to ensure continued observation and future paths for new dis-coveries of the ocean biosphere.

2. Data and methods

2.1. SeaWiFS observations of remote sensing reflectance

SeaWiFS was a polar-orbiting, sun-synchronous, crosstrack-scanninginstrument with six visible “ocean color” wavebands and two near-

Table 1Validation for bio-optical models using in situ and SeaWiFS satellite remote sensingreflectance observations.

Parameter N Intercept Slope R2 RMS deviation Bias

Validation using in situ remote sensing reflectance observationsChlOC4 2687 −0.008 0.975 0.859 0.275 −0.005ChlGSM 2528 −0.058 1.040 0.850 0.292 −0.063CDM 878 0.097 1.115 0.850 0.256 −0.032BBP 314 0.424 1.160 0.632 0.197 0.014

Validation using SeaWiFS remote sensing reflectance observationsChlOC4 1543 −0.011 0.981 0.795 0.310 −0.008ChlGSM 1380 −0.175 0.958 0.774 0.363 −0.165CDM 484 0.239 1.170 0.566 0.336 0.014BBP 201 0.047 1.048 0.531 0.210 −0.071

All comparisons aremade comparing log10-transformed variables. In situ validation is doneusing ocean remote sensing reflectance determinations from an updated version of theNOMAD data set (NOMADv2; following Werdell & Bailey, 2005) Matchups with satelliteremote sensing reflectance values are calculated using the single level-3 (9 km) pixelcontaining the observation for that day and the NOMADv2 field observations.

79D.A. Siegel et al. / Remote Sensing of Environment 135 (2013) 77–91

infrared (NIR) “atmospheric correction” bands (McClain et al., 2004).SeaWiFS collected imagery at a nominal 1-km spatial resolution (globallysubsampled to 4 km) and achieved near-complete global measurementcoverage every 2 days. Accurate ocean color retrievals are among themost difficult Earth-system satellite data to acquire, in part becauseonly ~10% of the signal measured by the satellite sensor at the top-of-atmosphere originates from the ocean (the remaining ~90% comesfrom the atmosphere and surface reflectance). SeaWiFS's NIR bandsprovide critical data for removing the atmospheric contribution to radi-ancesmeasured at the 6 ocean color bands. Sensor calibration and char-acterization (pre- and post-launch) and corrections to remove theeffects of scattering and absorption by the atmosphere and reflectionfrom the ocean surface have proven crucial for creating high qualitydeterminations of the spectral reflectance emanating from the ocean.

The SeaWiFS mission pioneered two system calibration proce-dures that helped it achieve the accuracy and stability required for cli-mate science (McClain, 2009; McClain et al., 2004; National ResearchCouncil, 2011). First, relative changes of the SeaWiFS's calibrationwere monitored by monthly viewing of the moon at constant phaseand each spectral band was characterized to ~0.1% over the mission'slifetime (Eplee et al., 2011). Second, absolute system calibrationfactors were determined through a vicarious calibration (Franz et al.,2007), which compares satellite observations to high-quality fieldmeasurements of ocean remote sensing reflectance propagated to thetop of the atmosphere. Following this approach, uncertainties in abso-lute calibration factors arewithin 0.3% for all visible bands and resultantretrievals of ocean remote sensing reflectance have uncertainties of ~3%for conditions similar to the field calibration site (see SupplementaryInformation section).

2.2. The SeaWiFS chlorophyll algorithm (OC4v6)

The ‘standard’ SeaWiFS chlorophyll product, ChlOC4, is based on anempirical, maximal band-ratio algorithm (O'Reilly et al., 1998) usingremote sensing reflectance measurements at 443, 490, 510 and555 nm. This algorithm, OC4v6, empirically relates the maximum ofthree remote sensing reflectance band-ratios, 443/555, 490/555 and510/555, to the chlorophyll concentration (see http://oceancolor.gsfc.nasa.gov/REPROCESSING/R2009/ocv6/). For low chlorophyll wa-ters, the 443/555 ratio will be the maximum ratio used, whereas foreutrophic waters the 510/555 ratios will be selected. The maximumband-ratio approach proved to be very effective for accurately re-trieving chlorophyll concentrations over a very large dynamic range(e.g., O'Reilly et al., 1998). Nearly 2000 paired observations of ocean re-mote sensing reflectance and chlorophyll concentration from version 2of the NASA bio-Optical Marine Data set (NOMADv2) archive of bio-optical fieldmeasurements (Werdell & Bailey, 2005)were used to derivethe OC4v6 algorithm.

Performance of the OC4v6 algorithm is excellent (Table 1) whenapplied to the NOMADv2 data set of ocean remote sensing reflectancespectra and compared with the in situ chlorophyll measurements(http://seabass.gsfc.nasa.gov/seabasscgi/nomad.cgi). This in situ datacomparison can be thought of as measuring the inherent variabilityin the OC4v6 algorithm, as the NOMADv2 observations comprisednearly all of the data used to determine the OC4v6 model coefficients.

SeaWiFS remote sensing reflectance spectra can be also used to as-sess the performance of the OC4v6 algorithm. A matchup data set ofSeaWiFS remote sensing reflectance spectra and in situ observations(largely from NOMADv2) is constructed using the single level-3pixel (~9 km) containing the in situ observation for that day. The per-formance of the ChlOC4 algorithm is again excellent (Table 1); with anR2 value of 0.795, slopes approaching unity, and a log-transformedintercept near zero. Similar validation results for ChlOC4 are found inthe analyses conducted by the NASA Ocean Biology ProcessingGroup following procedures laid out by Bailey & Werdell, 2006.Thus, the SeaWiFS band ratio chlorophyll algorithm (OC4v6) has

excellent performance when compared with the NOMADv2 in situchlorophyll data, both using in situ reflectance determinations andSeaWiFS reflectances.

2.3. The Garver–Siegel–Maritorena ocean color algorithm andits interpretation

Advances in ocean optics modeling have yielded tools to decom-pose satellite ocean color signals into multiple, specific optical contri-butions (e.g., IOCCG, 2006). Here, the Garver–Siegel–Maritorenaalgorithm (GSM; Maritorena et al., 2002, 2010) is applied to SeaWiFSocean reflectance data to estimate chlorophyll concentration (ChlGSM),the combined absorption at 443 nm by CDOM and detrital particulates(CDM), and the particulate backscattering coefficient at 443 nm (BBP).The GSM model retrieves ChlGSM, CDM and BBP simultaneously byperforming a non-linear least-squares minimization between the satel-lite andmodeled reflectance for all available visible bands.Model param-eters were evaluated through a constrained optimization using observedremote sensing reflectance, chlorophyll, CDM and BBP determinationsmostly from open ocean, non-polar environments (Maritorena et al.,2002). Our goal in applying theGSMmodel is to elucidate phytoplanktonchlorophyll changes independently from CDM (Siegel et al., 2005a),while providing a proxy for phytoplankton carbon biomass using BBP(Behrenfeld et al., 2005).

The performance of the GSM algorithm can be evaluated usingboth in situ and satellite data. Using the NOMADv2 in situ reflectancespectra, the GSM retrievals of chlorophyll concentrations are excel-lent, with performance metrics that are nearly equivalent to thoseof the operational SeaWiFS algorithm (Table 1). This is a noteworthyaccomplishment considering that the GSM parameters are optimizedto simultaneously retrieve values of ChlGSM, BBP and CDM, not justchlorophyll concentration (Maritorena et al., 2002). Further, veryfew of the observations used by Maritorena et al. (2002) to optimizethe GSM model are contained in the NOMADv2 data set (44 of 1024).The GSM algorithm also performs very well in terms of CDM re-trievals based on the NOMADv2 data set, while uncertainties aresomewhat greater for the BBP product (although there are also farfewer field BBP data available; Table 1).

End-to-end performance of the ChlGSM retrievals for SeaWiFS(R2 = 0.774) is nearly as good as the OC4v6 retrievals although asmall bias is observed (Table 1). However, GSM model performancewith the satellite matchup data for the two other data products isnot as good as found using in situ data. Differences in performancebetween the OC4v6 and the GSM model are expected because theGSM model works on absolute reflectance values, while OC4v6 algo-rithm uses ratios of reflectance determinations. Hence, GSM retrievalswill be more sensitive to noise in the reflectance retrievals than

http://oceancolor.gsfc.nasa.gov/REPROCESSING/R2009/ocv6/

http://oceancolor.gsfc.nasa.gov/REPROCESSING/R2009/ocv6/

http://seabass.gsfc.nasa.gov/seabasscgi/nomad.cgi


band-ratio based algorithms. Satellite spectral reflectance data aregenerally noisier than in situ measurements, particularly in areaswhere atmospheric correction is problematic (e.g., Gordon, 1997) orin the open ocean where the selection of digitization levels of theNIR bands for SeaWiFS creates noise in open ocean chlorophyll re-trievals (Hu et al., 2001). These issues clearly affect the quality of GSMretrievals. Overall the GSM model applied to the SeaWiFS reflectanceobservations performs well when compared with the NOMADv2matchup data set, particularly for ChlGSM (Table 1).

Retrievals of colored dissolved and detrital organic material(CDM) are dominated by the dissolved fraction (CDOM) in openocean waters (Fig. 1; Siegel et al., 2002b; Nelson et al., 2010; Nelson& Siegel, 2013). The mean value for the ratio of the detrital particulateabsorption coefficient at 443 nm to CDM from the global, open ocean,in situ data shown in Fig. 1 is 16.5% (std. dev. = 13.1%; N = 371).Hence variations in retrieved CDM values will be dominated bychanges in the CDOM concentration.

The particulate backscattering coefficient (BBP) is used as a proxyfor changes in phytoplankton carbon biomass as there are very fewdirect measurements of phytoplankton carbon concentrations thatcan be used to derive empirical algorithms (e.g., Behrenfeld et al.,2005; Westberry et al., 2008). Several observations support the appli-cation of the BBP proxy for phytoplankton carbon biomass. First, arelated optical property, the particulate beam attenuation coefficientat 660 nm (cp) is most sensitive to particles in the size range~0.5–20 μm (Stramski & Kiefer, 1991), a window which overlapsmost phytoplankton found in the open ocean. Changes in cp have re-peatedly been shown to track phytoplankton growth and abundanceover diel cycles (Durand & Olson, 1996; Gernez et al., 2011; Green et al.,2003; Siegel et al., 1989) and values of the chlorophyll normalized cp(cp:Chl) are strongly correlated with phytoplankton photosyntheticindices (Behrenfeld & Boss, 2003, 2006). Thus, there is a strong bodyof evidence linking changes in cp and phytoplankton biomass. More re-cently, regional observations have shown relationships between cp andBBP (Antoine et al., 2011; Dall'Olmo et al., 2009, 2011; Westberry et al.,2010). Thus, changes in BBP in the open ocean should be strongly linkedto changes in phytoplankton biomass (see also Behrenfeld et al., 2005;Westberry et al., 2008; Huot et al., 2008). Such relationships will notnecessarily apply in coastal regions where there is substantial inputsof terrigenous material or for regions of the world ocean where partic-ulate backscattering is influenced by coccolithophore abundances(e.g., Balch et al., 2011). However for much of the world ocean, BBPwill be a useful proxy for phytoplankton biomass.

2.4. Specific data and analyses employed

Monthly level-3 imagery (~9 km) from the SeaWiFS mission isused for this analysis (version 2010.0; http://oceancolor.gsfc.nasa.gov). Monthly images are used from October 1997 to November2010, although after January 2008 there are several months of miss-ing observations due to various spacecraft and communication issues.Data products include ocean remote sensing reflectance at 412, 443,490, 510, 555 and 670 nm, the operational chlorophyll concentration(ChlOC4), and the daily incident flux of photosynthetically availableradiation (PAR). Sea surface temperature (SST) fields were availablefrom the Advanced Very High Resolution Radiometer (AVHRR) Path-finder version 5.2 at a 4 km spatial resolution. Only nighttime SSTimagery is used to minimize effects of diurnal heating.

All ocean color satellite data were also mapped and averaged into1° by 1° latitude/longitude bins prior to analysis defining the regionalbins used in this analysis. All algorithms were applied to the level-3binned ocean remote sensing reflectance spectra before mappingand averaging into 1° regional bins. Values of ChlOC4, ChlGSM, andCDM were log-transformed before averaging following Campbell(1995). Estimates of median mixed layer growth irradiance, Ig, aredetermined as PAR exp(−KPAR MLD/2), where KPAR follows Morel

et al. (2007) and MLD determinations are from the Fleet NumericalOceanography Center model output (Clancy & Sadler, 1992). Monthlyanomalies were calculated as the difference between monthly meanvalues for each regional bin and the mission-long monthly mean forthat bin. Global aggregate anomalies for the warm/cool oceans,based upon the locations of the mission mean 15 °C SST isotherm,were calculated from the regional binned anomalies after accountingfor differences in the latitudes of the regional bins (Behrenfeld et al.,2006). All trend calculations were made by type 1 linear regressionand only significant (by the 95% c.i.) trends are shown. Trends inlog-transformed variables are presented in % per year, noting thatthe change in a natural log transformed variable is equivalent to thenormalized rate of change (Campbell, 1995).

3. Results

3.1. Observations using the SeaWiFS operational chlorophyll products

The mean band ratio chlorophyll concentration (ChlOC4) for theSeaWiFS mission illustrates several important features of the globalocean biosphere (Fig. 2). First, the spatial distribution of chlorophylllargely reflects broad-scale patterns in wind-driven vertical transportsof nutrients (e.g., McClain, 2009; Sverdrup, 1955; Yoder et al., 1993).Prominent regions of low chlorophyll are found in the subtropicalgyres, where downwelling Ekman pumping prevails. In contrast, re-gions of deep seasonalmixing (e.g., high latitudes) or persistent upwell-ing (e.g., along equator, Arabian Sea, eastern boundary currents) exhibitelevated chlorophyll levels (Fig. 2). Second, spatial differences in meanChlOC4 values span nearly 2 orders of magnitude — from 0.03 to>1 mg m−3 with a near log-normal distribution. Third, the 15 °C SSTmean isotherm (black line in Fig. 2) effectively delineates the produc-tive, high-chlorophyll sub-polar and polar regions from the nutrient-impoverished, low-chlorophyll regions of the tropics and subtropics(as in Behrenfeld et al., 2006).

SeaWiFS was launched during the peak of a strong El Niño and therelaxation to the cooler, La Niña phase was the primary driver forobserved global increases in surface chlorophyll during the earlyyears of the mission (e.g., Behrenfeld et al., 2001; Yoder & Kennelly,2003). Following the La Niña, mean ocean temperatures oscillatedbetween minor warming and cooling periods, while chlorophylllevels tracked fluctuations in SST (e.g., Behrenfeld et al., 2006;Gregg et al., 2005; Martinez et al., 2009). Relationships between SSTand ChlOC4 are easily seen in the monthly anomalies for the threeglobal aggregates: the cool northern hemisphere (NH) region (meanSST b 15 °C; Fig. 3a), the warm, permanently-stratified ocean (meanSST > 15 °C; Fig. 3b), and the cool southern hemisphere (SH)(mean SST b 15 °C; Fig. 3c). Note that the SST axis for each panel inFig. 3 is inverted to illustrate the correspondence between ChlOC4and -SST monthly anomalies. All three global aggregates exhibitstrong interannual patterns, with the 1998–99 El Niño–La Niña tran-sition from warm to cool SST values (and from low to high ChlOC4 re-trievals) being most clearly seen in the warm ocean aggregate. In factfor all three global aggregate regions, monthly anomalies for ChlOC4are significantly correlated with changes in SST, where an inverserelationship is found for the warm and cool NH oceans but a positiverelationship is found for the cool SH ocean (Table 2). Note that SST isconsidered here as a proxy for changes in physical ocean conditions,where anomalously warm SST's are associated with decreasing mixedlayer depth, suppressed surface nutrient input, and higher averagelight exposure for the near-surface mixed layer.

Highly significant, increasing linear trends in SST are found for allthree global aggregates, varying from0.015 to 0.035 °C per year andwiththe greatest increases occurring in the cool NH ocean (Fig. 3; Table 3). Atthe same time, statistically significant decreasing trends in ChlOC4 arefound only over the warm ocean aggregate (−0.18% year−1) while

http://oceancolor.gsfc.nasa.gov

http://oceancolor.gsfc.nasa.gov

Fig. 2. Mission mean chlorophyll concentration from SeaWiFS from August 1, 1997 to December 14, 2010. The OC4v6 band-ratio Chl algorithm (ChlOC4) is used. Mean values oflog(ChlOC4) are calculated for 1° bins in latitude and longitude over the global ocean. Units are log10(mg m−3). The location of the mean SST = 15 °C isotherm is shown as theblack lines. This is an update of figure 3.36a from Siegel et al. (2011) using the 2010.0 version of the SeaWiFS data set.


increasing trends are observed in the cool SH (trend = 0.83% year−1)and insignificant trends for the cool NH aggregate (Table 3).

The global aggregates (Fig. 3) provide an integrated view of net re-lationships between ocean biology and physical ocean conditions, butthey do not illustrate these processes on regional scales (here definedas 1° bins in latitude and longitude). The correlation between ChlOC4and SST regional scale anomalies shows extensive regions of bothnegative and positive correlations (Fig. 4a). The dominant pattern inthe warm stratified ocean is an inverse correlation between ChlOC4and SST anomalies (blue in Fig. 4a), with only small regions of posi-tive correlation (red regions in Fig. 4a; e.g., west of Central America,between Madagascar and Western Australia). In contrast, the coolNH and SH waters show coherent regions of both positive and nega-tive correlations, although much of the two cool global ocean aggre-gates show no significant correlation. Again, the 15 °C SST isotherm(black lines in Fig. 4a) approximately delineates the warm oceanwhere predominantly inverse correlations are found from the restof the oceans.

Regional temporal trends of log(ChlOC4) over the 13 years of theSeaWiFS record show large areas of both increasing (red areas inFig. 4b) and decreasing (blue areas) values. These regional trend pat-terns show some resemblance with the spatial distribution of SSTtrends (Fig. 4c) and are consistent with those analyzed by Vantrepotte& Mélin, 2009 using the first 10 years of the SeaWiFS data record(their Fig. 6b). The two trend estimates show aweak significant, inversecorrelation from the warm ocean aggregate (R2 = 0.077, p b 0.001),but this relationship explains a very small amount of variability. Forthe two cool ocean basins, correspondence between the regional scaletrends is even more tenuous. Clearly, the aggregate trends of Fig. 3have many underlying regional nuances, with magnitudes ten-foldgreater than the global summaries. These regional areas of large decadaltrends in ChlOC4 (absolute values > 30% over a decade; Fig. 4b) are like-ly caused by migration of boundaries between bio-optical provinces inresponse to regional changes in physical ocean climate. Clearly, the spa-tial variability in regional scale trends in ChlOC4 (Fig. 4b) demonstratesthe challenges in assessing global ocean biosphere changes from asparse array of observational assets.

3.2. CDM and the remote sensing of chlorophyll

The SeaWiFS operational chlorophyll products provide a first-orderview of phytoplankton dynamics on regional to global scales. However,the SeaWiFS algorithm assumes a fixed relationship among opticalproperties, which may not hold in all conditions (e.g., Dierssen, 2010;Siegel et al., 2002b, 2005b; Szeto et al., 2011). To evaluate the impor-tance of independence among optical properties, we use the GSM

ocean color algorithm, which enables phytoplankton chlorophyllchanges to be quantified separately from absorption due to CDOM anddetrital particles, while providing an independent measure of phyto-plankton carbon biomass via the particulate backscattering coefficient.The GSM data products are presented here to provide greater insightsinto the processes regulating phytoplankton dynamics.

The significance of variable CDM absorption on global chlorophyllretrievals can be illustrated using the normalized difference between theband ratio and the GSM-derived chlorophyll concentrations, ΔChlnorm(Fig. 5a; after Siegel et al., 2005b). The mean ΔChlnorm distributionshows that ChlOC4 retrievals are ~20% lower than ChlGSM within thesubtropical gyres, while ChlOC4 retrievals are considerably higher thanChlGSM values in the subpolar oceans, with differences of more than60% occurring in the northern hemisphere. Again, the 15 °C SST isothermeffectively delineates these two regions (black lines in Fig. 5a).

Many of the differences between the chlorophyll retrievals can beexplained by the presence of CDM as retrieved by the GSMmodel. Themean distribution of the fraction of non-water light absorption at443 nm regulated by CDM (%CDM) varies from ~40 to more than70% (Fig. 5b). As demonstrated previously, values of %CDM are lowin the subtropical gyres and higher at higher latitudes (Siegel et al.,2002b, 2005b). Importantly, the mean spatial distribution of ΔChlnormlooks very similar to the mean %CDM distribution, with their corre-spondence being particularly dramatic at high northern latitudesand the correlation between the two mean spatial patterns is highlysignificant (R = +0.738; p b 0.001). On a regional scale, temporalchanges in ΔChlnorm are also highly correlated with changes in%CDM, revealing a strong positive correlation throughout the warmocean (where mean SST > 15 °C; Fig. 5c). Correlations between re-gional changes in %CDM and ΔChlnorm are not as strong in either ofthe cool ocean domains. Together, these data confirm the need to ac-count for CDM influences on assessments of phytoplankton dynamicsfrom satellite ocean color observations (e.g., Sauer et al., 2012; Siegelet al., 2005b; Szeto et al., 2011).

3.3. Global to regional scale variability of CDM-partitioned chlorophyll

The GSMmodel provides chlorophyll determinations independentfrom the competing influences of CDM. The correspondence betweeninterannual trends in monthly anomalies in ChlGSM and CDM and seasurface temperature (SST) are shown for the cool NH aggregate(Fig. 6a, b), the warm, permanently-stratified ocean (Fig. 6c, d), andthe cool SH ocean (Fig. 6e, f). As before, all three global aggregatesexhibit strong interannual patterns, with the 1998–99 El Niño–LaNiña transition from warm to cool SST values (and from low to highChlGSM retrievals) beingmost clearly seen in thewarm ocean aggregate.

a

b

c

Fig. 3. Time series of ChlOC4 and sea surface temperature (SST) monthly standardizedanomalies (z-scores) for three global regions delineated by the mean SST isothermfor a) the cool (mean SST b 15 °C) northern hemisphere (NH) aggregate, b) the warm,permanently-stratified ocean aggregate (mean SST > 15 °C) and c) the cool southernhemisphere (SH) (mean SST b 15 °C). Anomalies are constructed by first removing themonthly mean value for each 1 degree bin of each property and then aggregating the re-gional, monthly anomalies into global aggregates. Statistics of the correlations betweenlog(ChlOC4) and SSTmonthly anomalies are given in Table 2 andmission length trend anal-yses are presented in Table 3. This is an update of figure 3.37 from Siegel et al. (2011) usingthe 2010.0 version of the SeaWiFS data set and the AVHRR Pathfinder v5.2 data set.

Table 2Correlations coefficients (R) between time series of monthly anomalies of SST andSeaWiFS retrievals for the three global aggregates.

ChlOC4 ChlGSM CDM

Region R p-Value R p-Value R p-Value

SST b 15 °C NH −0.2496 0.0019 +0.0566 0.4879 −0.2967 0.0002SST > 15 °C −0.5539 b0.0001 −0.5427 b0.0001 −0.7620 b0.0001SST b 15 °C SH +0.3013 0.0002 +0.3772 b0.0001 +0.0480 0.5599

Correlation values are bolded for p values less than 0.05.

Table 3Trends in the three SeaWiFS retrievals and SST vs. time over the SeaWiFS mission.

Region Slope 95% confidenceinterval for Slope

R2 p-Value

SST (slope units = °C/year)SST b 15 °C NH 0.0352 0.0229 0.0476 0.1684 b0.0001SST > 15 °C 0.0152 0.0101 0.0202 0.1824 b0.0001SST b 15 °C SH 0.0288 0.0234 0.0341 0.4165 b0.0001

ChlOC4 (slope units = %/year)SST b 15 °C NH 0.0373 −0.2702 0.3449 0.0004 0.8107SST > 15 °C −0.1758 −0.3061 −0.0455 0.0453 0.0085SST b 15 °C SH 0.8300 0.6217 1.0383 0.2924 b0.0001

ChlGSM (slope units = %/year)SST b 15 °C NH 0.7960 0.4605 1.1314 0.1278 b0.0001SST > 15 °C −0.0803 −0.1696 0.0089 0.0206 0.0774SST b 15 °C SH 1.0981 0.8771 1.3192 0.3911 b0.0001

CDM (slope units = %/year)SST b 15 °C NH −0.5587 −0.8689 −0.2484 0.0778 0.0005SST > 15 °C −0.3136 −0.4190 −0.2082 0.1871 b0.0001SST b 15 °C SH −0.0428 −0.1800 0.0945 0.0025 0.5391

Correlation values are bolded for p values less than 0.01.


Further, changes in both ChlGSM and CDM show correspondence withchanges in SST for the other two global aggregates (Fig. 6).

The covariation of both ChlGSM and CDMwith SST is not unexpected(e.g., Behrenfeld et al., 2005, 2008; Nelson & Siegel, 2013; Siegel et al.,2005a; Swan et al., 2009). As the physical ocean decreases mixedlayer depth and inhibits vertical mixing (leading to elevated SST), sub-surface nutrient fluxes are suppressed and average light levels for thenear-surface mixed layer increases. These factors will reduce retrievalsof both ChlGSM and CDM through the processes of phytoplanktonphysiological acclimation and photobleaching of CDOM, respectively(e.g., Nelson & Siegel, 2013; Siegel et al., 2005a). When the mixedlayer depth increases and vertical mixing intensifies (leading to reduc-tions in SST), new nutrients and CDOM are transported into the eupho-tic zone leading to strong increases of both ChlGSM and CDM.

The correlations found between interannual anomalies in CDM andSST for the permanently stratified oceans (R = −0.762; p b 0.001;

Fig. 6d) are the strongest found in this study (Table 2). Strong negativecorrelations are also foundbetween CDMand SST for the cool NH region(R = −0.297; p b 0.001; Fig. 6b). In both cases, warming is associatedwith decreasing CDM and cooling with increasing CDM. This is consis-tent with greater photobleaching of CDOM in shallower mixed layersdue to enhanced average light levels (e.g., Nelson & Siegel, 2013;Siegel et al., 2002b, 2005a; Swan et al., 2009, 2012). The strongest rela-tionships between SST and ChlGSM anomalies are found for the perma-nently stratified oceans (R = −0.543; p b 0.001; Fig. 6c) and the coolSH region (R = +0.377; p b 0.001; Fig. 6e). In the warm ocean aggre-gate, ChlGSM and SST changes are again inversely related, which is con-sistent with physiological responses of phytoplankton to increasinglight and decreasing nutrients (Behrenfeld et al., 2008). In the cool SH,chlorophyll anomalies are positively correlated with SST changes, indi-cating that a warming surface layer is associated with improved condi-tions for biomass accumulation (either reflecting increased availabilityof a limiting resource for growth or ecosystem-level changes in phyto-plankton growth-loss relationships; Doney, 2006; Behrenfeld, 2010).

As shown previously, highly significant, increasing SST trendsare observed for all three global aggregates over the SeaWiFS era(Table 3). Statistically significant increases in ChlGSM are observed inthe cool NH (trend = 0.80% year−1; Fig. 6a) and the cool SH oceans(trend = 1.10% year−1; Fig. 6e), while long-term trends for ChlGSMover the warm ocean aggregate were statistically insignificant(Fig. 6c; Table 3). Conversely, CDM trends over the SeaWiFS eradecreased within both the warm ocean (−0.31% year−1; Fig. 6d)and the cool NH ocean region (−0.56% year−1; Fig. 6f), while thetrends in CDM over time were insignificant for the cool SH aggregate(Table 3).

Again, regional-scale correlations between ChlGSM and SST anomaliesshow large regions of both negative and positive correlations (Fig. 7a)and look very similar (although slightly muted) to linear correlation

a

b

c

Fig. 4. Regional-scale spatial patterns in the relationship between log(ChlOC4) and SST and their trends over time. a) Values of the regression coefficient (R) between log(ChlOC4) andSST, b) linear ln(ChlOC4) trend (in % year−1) and c) linear SST trend (in °C year−1). Trends are calculated over the entire SeaWiFS mission duration. Statistics calculated for each 1°latitude-longitude bin. Only significant (at the 95% c.i.) correlation or trend determinations are plotted. The mean SST = 15 °C isotherm is shown as the black lines in each panel.This is an update of figure 3.38 from Siegel et al. (2011) using the 2010.0 version of the SeaWiFS data set and the AVHRR Pathfinder v5.2 data set. (For interpretation of the references tocolor in this figure legend, the reader is referred to the web version of this article.)


patterns seen between ChlOC4 and SST (Fig. 4a). The dominant pattern inthe warm stratified ocean is an extensive inverse correlation betweenChlGSM and SST anomalies particularly in the tropical oceans and forthe northern boundary of the subtropical gyres while correlation valuesare much less coherent in for the cool NH and SH waters (Fig. 7a). Aswith ChlOC4, linear trends of log(ChlGSM) over the SeaWiFS record showlarge areas of both increasing (red areas in Fig. 7b) and decreasing(blue areas) values.

The distribution of the mission-long temporal trends for log(CDM)(Fig. 7c) shows many of the same spatial patterns as seen forlog(ChlOC4) and log(ChlGSM) (Figs. 4b and 7b). However the spatialpatterns for log(CDM) trends show more extensive regions of de-creasing values compared with both chlorophyll indices supportingthe global scale aggregate results (Table 3). The correlations betweenthe log(ChlGSM) and log(CDM) regional trends are very strong overthe entire domain (R2 = 0.650; p b 0.001).

3.4. Roles of physiology vs. biomass accumulation in regulatingchlorophyll variability

The large-scale distribution of phytoplankton chlorophyll concen-tration reflects largely patterns in nutrient supply and thereby differ-ences in phytoplankton abundances (Fig. 1). Phytoplankton can alsoadjust their intracellular chlorophyll concentrations in response totheir light and nutrient environs (e.g., Behrenfeld et al., 2002, 2005;Laws & Bannister, 1980). An important question is how much of thephytoplankton chlorophyll variability (Fig. 8a) is reflective of changesin phytoplankton biomass versus cellular chlorophyll changes in re-sponse to light (e.g., Behrenfeld et al., 2002, 2005, 2008; Siegel et al.,2005a;Westberry et al., 2008). To address this question, we statisticallypartition the variability in theGSMretrieved chlorophyll concentrations(Fig. 8a) using remote sensing proxies for phytoplankton biomass andthe chlorophyll to carbon ratio. The GSM-retrieved BBP estimate is

a

b

c

Fig. 5. The importance of CDM on retrieval phytoplankton chlorophyll retrievals from satellite ocean color observations. a) The mission mean normalized chlorophyll concentrationanomaly, ΔChlnorm. Values of ΔChlnorm are calculated as the normalized difference between ChlOC4 and ChlGSM (ΔChlnorm = 100 ∗ (ChlOC4 − ChlGSM) / ChlGSM; Siegel et al., 2005b).b) The mission mean fraction of non-water light absorption at 443 nm that can be attributed to CDM, %CDM (Siegel et al., 2005b). Values of %CDM are calculated as100 ∗ CDM / (CDM + aph(440;ChlGSM)) where aph(440;ChlGSM) = 0.0378 (ChlGSM)0.627 (Bricaud et al., 1998). The Bricaud et al. (1998) aph(440) model is used in the %CDM definitionbecause the GSM model was constructed to estimate values of ChlGSM and not aph(440) (Maritorena et al., 2002). c) The regional-scale distribution of the correlation coefficient (R) forthe relationship between ΔChlnorm and %CDM. Only significant (at the 95% c.i.) correlation values are plotted. The mean SST = 15 °C isotherm is shown as the black lines. This updatesfigures from Siegel et al. (2005a) and Siegel et al. (2005b) using the 2010.0 version of the SeaWiFS data set.


used to account for changes in phytoplankton biomass (Section 2.3).Changes in the regional chlorophyll to carbon ratios due to changinglight conditions are modeled as exp(−3 Ig), where Ig is the mediangrowth irradiance of themixed layer (Section 2.4). The photoacclimationterm used (exp(−3 Ig)) was originally derived from fieldmeasurementsfrom the Atlantic Ocean (Behrenfeld et al., 2002). It was later shownby Behrenfeld et al. (2005) to effectively account for satellite-observedregional relationships between Ig and chlorophyll to carbon ratios. Fur-ther, Westberry et al. (2008) reported the same photoacclimationresponse for nutrient-replete phytoplankton populations using globalsatellite data. These findings support this parameterization of thephotoacclimation response, but further validation is needed and willrequire routinefieldmeasurements of phytoplankton carbon and chloro-phyll (e.g., Graff et al., 2012). The impact of the photoacclimation

response on chlorophyll concentrations is therefore the product ofexp(−3 Ig) and BBP and the variability in ChlGSM can be modeled as

ChlGSM ¼ fbioBBPþ fphysBBP exp −3 Ig� ��

ð1Þ

where fbio and fphys are regression coefficients that quantify the relativeimportance of biomass and photoacclimation changes on variations ofthe satellite sensed chlorophyll concentration. All terms in Eq. (1) arestandardized (i.e., zeromean, unit variance) so the derived linear regres-sion coefficient values provide, for each regional (1°) bin, a quantificationof the relative contributions made by the two processes.

The global distribution of fbio (Fig. 8b) shows dominance (i.e., elevatedvalues ≥ 0.7) throughout the high latitude oceans (poleward of the 15 °C

a b

c d

e f

Fig. 6. Relationships between ChlGSM and CDM standardized monthly anomalies (z-scores) with changes in sea surface temperature (SST) for three global regions delineated by themean SST isotherm. The top panels show the cool (mean SST b 15 °C) northern hemisphere (NH) aggregate for a) log(ChlGSM) and b) log(CDOM). The middle panels show thewarm, permanently-stratified ocean aggregate (mean SST > 15 °C) for c) log(ChlGSM) and d) log(CDOM). The lower panels show the cool southern hemisphere (SH) (meanSST b 15 °C for e) log(ChlGSM) and f) log(CDOM). Anomalies are constructed by first removing the monthly mean value for each 1 degree bin of each property and then aggregatingthe regional, monthly anomalies into global aggregates. Statistics of the correlations between the log(ChlGSM) and log(CDOM) anomalies and the SST anomalies are given in Table 2and mission length trend analyses are presented in Table 3.


SST isotherm) and in areas of large-scale, persistent or seasonal upwelling(e.g., eastern equatorial Pacific, eastern boundary currents, Arabian Sea,etc.). These are the ocean regions where phytoplankton biomass changescovary with chlorophyll variability. For these regions changes in remotesensed chlorophyll concentrations will be a good proxy for assessingchanges in phytoplankton biomass. However throughout much of thesubtropical ocean, fphys dominates chlorophyll variability (Fig. 8c). Highvalues of fphys are also found in the central tropical Pacific Ocean and intheMediterranean Sea. These broad red areas in Fig. 8c correspond to re-gions where phytoplankton physiological responses to seasonal changesin light are the dominant cause of surface chlorophyll variability. Insome areas where fphys is dominant (>0.5), retrieved values of fbio arenegative (blue areas in Fig. 8b; e.g., western north Pacific and SouthAtlantic and South Pacific Subtropical Gyres). An inverse relationshipbetween ChlGSM and phytoplankton biomass was also found in manyof the same regions of the world's oceans (see figure 2 of Behrenfeldet al., 2005). There are also broad regions of the oceanwhere the regres-sion coefficients are not significantly different from zero (white areas inFigs. 8b and 8c). These regions largely coincide in regionswhere there is

little variability in chlorophyll concentrations (Fig. 8a). The proxies forphytoplankton carbon concentration and the chlorophyll to carbonratio used in this analysis are at a nascent stage of development. Yetthey demonstrate that satellite-retrieved chlorophyll concentrationscontain information about changes in both phytoplankton biomass andits physiological state and that chlorophyll concentrations alone areoften an inappropriate proxy for phytoplankton abundance variations.

4. Discussion and conclusions

Our analysis of regional- to global-scale phytoplankton dynamicsusing SeaWiFS ocean color observations illustrates several importantresults about the interpretation of changes in phytoplankton propertiesassessed from satellite ocean color observations. First, the global oceanbiosphere is too complex to be described by a single index, such as chlo-rophyll concentration. Second, this complexity has important bearingson how climate-relevant trends in the ocean biosphere are assessed.Third, the simple comparison of global and regional scale trends illus-trates the difficulty in extrapolating from the time series sampling of a

a

b

c

Fig. 7. Regional-scale spatial patterns in the relationship between log(ChlGSM) and SST and the trends in log(ChlGSM) over time. a) Values of the regression coefficient (R) betweenlog(ChlGSM) and SST, b) linear log(ChlGSM) trend (in % year−1) and c) linear log(CDM) trend (in % year−1). Trends are calculated over the entire SeaWiFS mission duration. Statisticscalculated for each 1° latitude-longitude bin. Only significant (at the 95% c.i.) correlation or trend determinations are plotted. The mean SST = 15 °C isotherm is shown as the blacklines in each panel.


relatively few locations to estimate global trends. Last, the SeaWiFSmis-sion provided a stable and traceable source of ocean remote sensing re-flectance spectra, which were critical for correctly analyzing temporalchanges in global phytoplankton dynamics presented here. In thefollowing, we discuss 1) the temporal trends in the ocean biospherefound on regional to global scales from the SeaWiFS mission and theirimplications for the assessment of trends on decadal to centennial timescales, 2) future approaches to the remote sensing of phytoplanktondynamics in a complex ocean, and 3) reflections for how the SeaWiFSmission provides a path for addressing these challenges in the future.

4.1. Global ocean biosphere trends over the SeaWiFS era and beyond

The present findings highlight several important points concerningthe assessment of long-term trends from satellite ocean color observa-tions. First, variability in the three global aggregates was dominated

by ocean climate variations on interannual time scales and not bylong-term trends over the 13-year lifespan of SeaWiFS. The strong cor-respondence between monthly anomalies in SST and ChlOC4 (and otherproperties) demonstrates the close interrelationship between oceanbiology and climate variations (Table 2; Figs. 3 & 6). In addition tothese dominant seasonal-to-interannual variations, significant 13-yearlinear trends are found (Table 3). These long-term trends, however,explain very little of the total observed variations. For example, themost significant statistically linear trend found for the warm ocean ag-gregate explained only ~19% of the observed variance (for log(CDOM);Fig. 6d; Table 3). It seems likely that a linear trend applied over a singlesatellitemissionwill be too short to properly assess long-term temporaltrends in the global ocean biosphere (e.g., Henson et al., 2010;Vantrepotte & Mélin, 2011).

The present results also provide information about the spatialscales for decadal trends in the global biosphere. A comparison of

a

b

c

Fig. 8. Chlorophyll variability and its relation to biomass and light-induced physiological changes. a) Standard deviation in log(ChlGSM) estimated over the SeaWiFS mission. b) Standardizedregression slopes for predicting ChlGSM as function of changes in biomass (fbio in Eq. 1) and c) and for changes in rates of photoacclimation (fphys). Values of the standardized regression slopesnear one indicate the dominance of that process on ChlGSM variability. Details ofmodel configuration are provided in the text. Only statistically significant (95% c.i.) slope values are shown. Themean SST = 15 °C isotherm is shown as the black lines in each panel. (For interpretation of the references to color in this figure legend, the reader is referred to theweb version of this article.)


the global (Fig. 3) and regional scale trends (Figs. 4b) illustrates thedifficulty in extrapolating from in situ sampling of a relatively few loca-tions to estimate global scale trends. Fixed-location time-seriesmeasure-ment programs, such as the Bermuda Atlantic Time series Station (BATS)and the Hawaii Ocean Time-series (HOT), are critical observatories of theprocesses driving changes in ocean biosphere. However, long-termtrends derived from them provide at best a single-point in a map likeFig. 4b. Clearly, data from many long-term monitoring sites – samplingboth areas of increasing and decreasing trends – are needed to create aglobal assessment of change.

The trends diagnosed here show both similarities and differenceswith previous studies of temporal trends found using SeaWiFS obser-vations. For example, Gregg et al. (2005) find a ~4% increase in globalchlorophyll concentrations over the first six years of the SeaWiFS re-cord (1998 to 2003), with the highest increases in coastal environ-ments. Contrasting this, Behrenfeld and Boss (2006) find decreasing

levels of global chlorophyll concentrations and rates of net primaryproduction (NPP; from 1999 to 2005) that follow strong increases inboth ChlOC4 and NPP during the transition in 1998 from El Niño to LaNina conditions. Vantrepotte and Mélin (2009) examine the first tenyears of the SeaWiFS record and find trends for biogeochemical prov-inces consistent with the present analysis, which is expected consider-ing the similarity in the periods studied. Methodology is likely to beimportant as well. The present study log-transforms the chlorophyllobservations before global scale anomalies are calculated while theGregg et al. (2005) study does not. Not assuming log-normality willlikely amplify the influence of the fewer, highly productive sites in glob-al summaries. Thus, record length, time period and assumptions in dataanalysis will all have important roles in quantifying trends.

Evaluating changes on time scales longer than a single mission re-quires the merging of data across multiple satellite platforms and/orthe inclusion of historical field observations. Several researchers


have link data from the Coastal Zone Color Scanner (CZCS) (1978–1985)with the SeaWiFS data record (e.g., Antoine et al., 2005; Gregg et al.,2003; Martinez et al., 2009). Gregg et al., 2003 find a global decline ofNPP of more than 6% between CZCS observations in the early 1980sand the first five years of the SeaWiFS record, withmost of the decreasesoccurring at high latitudes. Contrasting these results, Antoine et al.(2005) find a ~22% increase in global chlorophyll concentrations,where most of these increases are due to changes at low latitudes.These two analyses are based on fundamentally different satellite dataprocessing approaches and underscore the importance of consistencyof data processing and radiometric calibration standards across multiplemissions (McClain, 2009; National Research Council, 2011; Siegel &Franz, 2010).

Only field observations can currently be used to examine trends ofglobal chlorophyll concentrations on longer time scales. RecentlyBoyce et al. (2010) created a centennial-scale time series of oceanchlorophyll concentrations using both discrete measurements of near-surface chlorophyll concentrations and Secchi disk depth (a measureof water clarity). Using this synthesized data set, Boyce et al. (2010)found an overall chlorophyll decrease over the past century of ~1%per year (and decreasing trends were found for 6 of their 8 global sub-regions). This implies that over the past 100 years there has been a~63% decrease in global chlorophyll concentrations, which we deemunlikely. The present analysis shows a significant decrease in ChlOC4only for the warm ocean aggregate and decrease was a factor of fivesmaller than the Boyce et al.'s result (Table 3). Further, significantincreases in ChlOC4 are found for the two cold ocean aggregates. Thesatellite data sets discussed here and the Boyce et al.'s (2010) analysisare based on very differentmethods and do not evaluate the same peri-od of time. However the estimation of a 1% annual global decrease inchlorophyll concentrations is also inconsistent with the fact that therate of global SSTwarming has been greater over the past three decadesthan earlier in the past century (Rayner et al., 2003; Xue et al., 2011).

Resolution of the validity of Boyce et al.'s, (2010) conclusions is be-yond the scope of this contribution, although the present discussionprovides some useful insights. We note that in response to publishedcomments on their original paper, Boyce et al. (2011) state that “…our statistical models reproduced with high fidelity the well-knownseasonal cycles of Chl in different regions and demonstrated clearcoherence between Chl and leading climate indicators.” Existing satelliteocean color observations have proven quite capable to assess seasonalvariations (e.g., Siegel et al., 2002a; Yoder et al., 1993) and responses toclimate oscillations (e.g., Behrenfeld et al., 2006; Martinez et al., 2009).However the robust quantification of multi-decadal trends from thesedata sets has proven to be difficult at best as demonstrated by the strik-ingly different conclusions drawn by different analysts using the exactsame data (e.g., Antoine et al., 2005; Gregg et al., 2003). We speculatethat similar issues may be affecting the Boyce et al.'s (2010) analysiswhere their trends may be retrieved improperly due to the dominanceof the larger seasonal and climate oscillation signals and inherent noisein their synthesized data set. A likely source of noise in the Boyce etal.'s (2010) data set is the lack of uniformity (and statistical indepen-dence) of available historic observations of the global ocean (see Fig. 1of Boyce et al., 2010) as compared to the expected spatial scales oflong-term trends (Figs. 4b & 7b). Further research is needed to better un-derstand the role of sampling on the centennial-scale trends estimatedby Boyce et al. (2010).

This discussion raises several important points about the quantifica-tion of long-term trends of the global biosphere. A thirteen-year mis-sion is a short amount of time to detect and quantify trends in theglobal ocean biological properties and small details in periods chosenand/or analysis periods can impact the trends diagnosed. Longer timeseries require multiple satellite missions and the creation and mainte-nance of a multi-satellite ocean color time series will require the carefulbridging between existing and plannedmissions, including the trackingof appropriate reflectance targets such as the moon and the multiple

reprocessing of available data sets (e.g., McClain, 2009; NationalResearch Council, 2011; Siegel & Franz, 2010). The development ofa multi-satellite ocean color time series remains the most feasibleapproach for documenting regional- to global-scale variability of theocean biosphere (National Research Council, 2011). The present analy-sis also provides context for evaluating the suitability of historical obser-vations for detecting long-term trends in the global biosphere.

4.2. Remote sensing of an optically complex ocean

A major point of this paper is to show that the dynamics of theopen ocean ecosystems are simply too complex to be described by asingle index, such as chlorophyll concentration. Understanding thiscomplexity requires quantifying the variability in relevant ocean colorconstituents, such as chlorophyll, phytoplankton carbon and CDOM,and their responses to perturbations in the physical ocean environment.Further additional factors not considered here, such as the phytoplank-ton community composition and particle size distribution, need to beassessed in the future to understand the role of phytoplankton dynamicsin regulating important biogeochemical factors, including carbon exportand net community production.

The optically complex ocean has implications for the interpretationof long-term trends in global phytoplankton chlorophyll observations.For example, the warm ocean aggregate trends for CDM and ChlOC4 de-crease significantly over time while the linear trend for ChlGSM vs. timeis statistically insignificant (Table 3). Also as shown previously, there isa strong correspondence between CDM and ChlOC4 where increases in%CDM create an overestimate of band-ratio retrieved values of chloro-phyll concentrations (Fig. 5c). Both lines of evidence indicate that theobserved trend in ChlOC4 for the warm aggregate ocean is likely due tochanges in CDOM, which dominates the CDM signal, and not phyto-plankton chlorophyll concentrations.

This uncertainty in the assessment of long-term trends in oceanbiosphere points to the importance of understanding the cycling ofopen ocean CDOM (e.g., Nelson & Siegel, 2002, 2013). The seasonaldynamics of surface CDOM are thought to be controlled by the samephysical forcings that regulate chlorophyll concentrations; namely theseasonal mixing of subsurface CDOM stocks and net CDOM productionduring the spring phytoplankton bloom followed by photobleachingof mixed layer CDOM stocks in the summer and a regeneration ofCDOM in the lower euphotic zone (e.g., Nelson & Siegel, 2013; Nelsonet al., 1998; Siegel et al., 2002b). However CDOM is comprised of asuite of chromophores and changes in CDOMwill occur on a multitudeof time scales from the time course of an uptake experiment in the lab-oratory to the multi-decadal time scale of cycling in the deep ocean(e.g., Kitidis et al., 2006; Nelson & Siegel, 2002, 2013; Nelson et al.,2004, 2010; Swan et al., 2009, 2012; Yamashita & Tanoue, 2009). Aquantitative description of CDOM dynamics is sorely needed, but thishas yet to be developed. CDOM is important in its own rights, as it hasa dominant role in solar light absorption (Fig. 1) and its variabilityhelps regulate many biogeochemical processes (e.g., Millet et al.,2010; Toole et al., 2008; Zepp et al., 2007).

There are also physiological complexities to the global oceanbiosphere. We have shown that in highly productive regions whereblooms are common, changes in chlorophyll concentration are domi-nated by changes in phytoplankton biomass. However throughoutmuch of the subtropical oceans, chlorophyll variations provide infor-mation primarily on physiological responses to light (and nutrients)and not changes to biomass (Fig. 8c). This means that remote sensingretrievals of chlorophyll are not always a good proxy for phytoplanktonconcentrations. The present analysis of biomass variability is based onBBP determinations (Behrenfeld et al., 2005, 2008; Westberry et al.,2008). Improvements to this method require global in situ observationsof phytoplankton biomass, which are presently lacking. Clearly, routinedeterminations of phytoplankton carbon concentration are thefirst step


towards developingbettermethods for assessing phytoplankton carbonfrom satellite observations (e.g., Graff et al., 2012).

Advances are also needed to assess changes in phytoplanktoncommunity structure from satellite observations (e.g., Dierssen, 2010).Assessments of phytoplankton community structure have been madeby either diagnosing phytoplankton functional type or phytoplanktonsize class (e.g., Alvain et al., 2005; Brewin et al., 2011; Bricaud et al.,2012; Kostadinov et al., 2010; Mouw & Yoder, 2010). Again, availablefield data sets remain a major limitation and currently only chemotax-onomic analysis of phytoplankton pigment concentrations or particlesize distribution determinations are used as “ground truth”. Bothapproaches are limited in their ability to diagnose phytoplanktoncommunity structure information (e.g., Higgins et al., 2011; Kostadinovet al., 2010) and this is another area where advances in fieldmethodolo-gies are needed.

Progress in understanding the optically complex ocean requiresimprovements and expansion of field data sets to develop and testbio-optical algorithms as well as advancements in space technology. Allocean color algorithms are empirical and the availability of high qualityobservations that span the application space for these algorithms iscentral for these developments. Many suggestions for improvements infield data needed for bio-optical model development have been madealready. One important suggestion is refining the quality of existingdata sets, such as NOMADv2, which will enable researchers to choosethe highest quality or most extensive observations to develop and testtheir models.

Improvements in space hardware are also needed to best accountfor the complex ocean environment. Successful atmospheric correction isa critical part of satellite ocean color remote sensing (see Section 2.1).This gets harder as several important oceanic and atmospheric constitu-ents, such as CDOM and absorbing aerosol concentrations, have similaroptical signatures yet they need to be eliminated to prevent spurious in-terpretations of phytoplankton dynamics (e.g., Gordon, 1997). Additionalocean color bands in the near-ultraviolet portion of the spectrum wouldhelp separate CDOM from phytoplankton absorption and will assist incharacterizing absorbing aerosol loads, as both of these optical propertiesstrongly absorb in the near-UV. Another area where improvements inspace-based hardware could be useful is the assessment of phytoplank-ton community structure. High spectral resolution satellite observations(~5 nm) are a very promising technique that can be used to distinguishthe contributions of individual phytoplankton pigment concentrations(e.g., Bidigare et al., 1989; Bracher et al., 2009; Torrecilla et al., 2011).Many of these suggestions have been incorporated in the planning docu-ment for the Pre-Aerosol, Clouds, and ocean Ecosystem (PACE) MissionScience Definition Team (PACE SDT, 2012) as well as several other inter-national satellite ocean color mission plans.

4.3. SeaWiFS legacy and its use as a periscope to the future

The SeaWiFS mission provided the first consistent, long-term,well-sampled and stable estimates of ocean remote sensing reflec-tance spectra and these observations were central for the analysis oftemporal changes in regional to global phytoplankton dynamicspresented here. The loss of the SeaWiFS data stream marks the endof an observational time-series largely free of instrument artifactsand a critical record against which other satellite ocean color sensorshave been made useful (e.g., Jeong et al., 2011; Kwiatkowska et al.,2008; Meister et al., 2012). Establishing a quality replacement forSeaWiFS that will sustain, as well as advance, the ocean color timeseries is of paramount importance (National Research Council, 2011;Siegel & Franz, 2010).

Results of the present study also provide unique insights for futuresatellite ocean color mission capabilities. First as concluded above,measurements of ocean remote sensing reflectance spectra need to besustained over a long time (>10 years) and be traceable to known stan-dards. Temporal changes of the ocean biosphere, particularly trends,

can only be evaluated knowing the quality of the underlying data set.This requires a vicarious calibration capability including appropriatefield sampling and an independent assessment of satellite instrumentstability, such aswould be available from lunar viewing. Thiswill enablethe setting of satellite gains and characteristics as a function of time. It isimportant that this capability be in place beyond the planned missionlifetime so records from multiple missions can be stitched together tomake the multi-decadal data sets needed.

This discussion bears consideration of what is meant by a “satellitemission”, which should be more than just the hardware and softwarerequired to make a measurement from space. Many of the pointsmade in the discussion so far, such as the need for vicarious calibra-tion, field data for model development, and sequential reprocessingof the data stream beyond the lifetime of a mission, suggest that anintegrated approach is required. Furthermore many of the questionsand problems, including the assessments of phytoplankton physio-logical state, carbon export, net community production, air-sea CO2

exchanges, all require the integration of satellite observations withEarth systemmodels and the use of detailed laboratory investigationsthat guide the development of these models. Hence, we suggest thatthe development of a “satellite mission” includes all aspects of themission required to answer the science questions that motivated itsimplementation in orbit.

SeaWiFS was unique as it created a blueprint for how ocean bio-sphere observations should be made on decadal time scales (NationalResearch Council, 2011). As reported here, analysis of the SeaWiFSobservations also provides directions for future satellite ocean colorobservations. These advanced observational tools will likely enablenew discoveries of the structure and functioning of ocean ecosystems,which will in turn lead to improved understanding of our oceans andtheir wise stewardship.

Acknowledgments

The authors would like to acknowledge the NASA Ocean Biologyand Biogeochemistry program for its long-term support of satelliteocean color research and the Orbital Sciences Corporation and GeoEyewho were responsible for the launch, satellite integration and on-orbitmanagement the SeaWiFS mission. The authors would also like tothank the anonymous reviewers for their excellent and extremely helpfulcomments on our manuscript.

Appendix A. Supplementary data

Supplementary information to this article can be found online athttp://dx.doi.org/10.1016/j.rse.2013.03.025.

References

Alvain, S., Moulin, C., Dandonneau, Y., & Bréon, F. M. (2005). Remote sensing of phyto-plankton groups in case 1 waters from global SeaWiFS imagery. Deep-Sea ResearchPart I, 52, 1989–2004.

Antoine, D., Morel, A., Gordon, H. R., Banzon, V. F., & Evans, R. H. (2005). Bridging oceancolor observations of the 1980s and 2000s in search of long-term trends. Journal ofGeophysical Research, 110. http://dx.doi.org/10.1029/2004JC002620 (C06009).

Antoine, D., Siegel, D. A., Kostadinov, T., Maritorena, S., Nelson, N. B., Gentili, B., et al.(2011). Variability in optical particle backscattering in contrasting bio-optical oceanicregimes. Limnology and Oceanography, 56, 955–973.

Bailey, S. W., & Werdell, P. J. (2006). A multi-sensor approach for the on-orbit validationof ocean color satellite data products. Remote Sensing of Environment, 102, 12–23.

Balch, W. M., Drapeau, D. T., Bowler, B. C., Lyczskowski, E., Booth, E. S., & Alley, D.(2011). The contribution of coccolithophores to the optical and inorganic carbonbudgets during the Southern Ocean Gas Exchange Experiment: New evidence insupport of the “Great Calcite Belt” hypothesis. Journal of Geophysical Research,116. http://dx.doi.org/10.1029/2011JC006941.

Behrenfeld, M. J. (2010). Abandoning Sverdrup's critical depth hypothesis on phyto-plankton blooms. Ecology, 91, 977–989.

Behrenfeld, M. J., & Boss, E. (2003). The beam attenuation to chlorophyll ratio: An opticalindex of phytoplankton photoacclimation in the surface ocean? Deep-Sea ResearchPart I, 50, 1537–1549.


http://dx.doi.org/10.1029/2004JC002620

http://dx.doi.org/10.1029/2011JC006941


Behrenfeld, M. J., & Boss, E. (2006). Beam attenuation to chlorophyll concentration asalternative indices of phytoplankton biomass. Journal of Marine Research, 64,431–451.

Behrenfeld, M. J., Boss, E., Siegel, D. A., & Shea, D. M. (2005). Carbon-based ocean produc-tivity and phytoplankton physiology from space. Global Biogeochemical Cycles, 19.http://dx.doi.org/10.1029/2004GB002299.

Behrenfeld, M. J., Halsey, K., & Milligan, A. (2008). Evolved physiological responses ofphytoplankton to their integrated growth environment. Proceedings of the RoyalSociety B: Biological Sciences, 363. http://dx.doi.org/10.1098/rstb.2008.0019.

Behrenfeld, M. J., Marañón, E., Siegel, D. A., & Hooker, S. B. (2002). Photoacclimationand nutrient-based model of light-saturated photosynthesis for quantifying oceanicprimary production.Marine Ecology Progress Series, 228, 103–117.

Behrenfeld, M. J., O'Malley, R. T., Siegel, D. A., McClain, C. R., Sarmiento, J. L., Feldman, G.C., et al. (2006). Climate-driven trends in contemporary ocean productivity. Nature,444, 752–755.

Behrenfeld, M. J., Randerson, J. T., McClain, C. R., Feldman, G. C., Sietse, O. L., Tucker, C. J.,et al. (2001). Biospheric primary production during an ENSO transition. Science,291, 2594–2597.

Bidigare, R., Morrow, J., & Kiefer, D. (1989). Derivative analysis of spectral absorption byphotosynthetic pigments in the western Sargasso Sea. Journal of Marine Research, 47,323–341.

Bishop, J. K. B., Davis, R. E., & Sherman, J. T. (2002). Robotic observations of dust stormenhancement of carbon biomass in the North Pacific. Science, 298, 817–821.

Boyce, D. G., Lewis, M. L., & Worm, B. (2010). Global phytoplankton decline over thepast century. Nature, 466, 591–596.

Boyce, D. G., Lewis, M. L., &Worm, B. (2011). Is there a decline in marine phytoplankton?Reply. Nature, 472, E8–E9. http://dx.doi.org/10.1038/nature09953.

Bracher, A., Vountas, M., Dinter, T., Burrows, J. P., Röttgers, R., & Peeken, I. (2009).Quantitative observation of cyanobacteria and diatoms from space using PhytoDOASon SCIAMACHY data. Biogeosciences, 6, 751–764.

Brewin, R. J. W., Hartman-Mountford, N. J., Lavender, S. J., Raitsos, D. E., Hirata, T., Uitz, J.,et al. (2011). An intercomparison of bio-optical techniques for detecting dominantphytoplankton size class from satellite remote sensing. Remote Sensing of Environment,115, 325–339.

Bricaud, A., Ciotti, A. M., & Gentili, B. (2012). Spatial-temporal variations in phytoplanktonsize and colored detrital matter absorption at global and regional scales, as derivedfrom twelve years of SeaWiFS data (1998–2009). Global Biogeochemical Cycles,26. http://dx.doi.org/10.1029/2010GB003952.

Bricaud, A., Claustre, H., Ras, J., & Oubelkheir, K. (2004). Natural variability of phyto-plankton absorption in oceanic waters: Influence of the size structure of algalpopulations. Journal of Geophysical Research, 109. http://dx.doi.org/10.1029/2004JC002419 C11010.

Bricaud, A., Morel, A., Babin, M., Allali, K., & Claustre, H. (1998). Variations of lightabsorption by suspended particles with the chlorophyll a concentration in oceanic(case 1) waters: Analysis and implications for bio-optical models. Journal ofGeophysical Research, 103, 31,033–31,044.

Campbell, J. W. (1995). The lognormal distribution as a model for bio-optical variabilityin the sea. Journal of Geophysical Research, 100, 13237–13254.

Ciotti, A. M., Lewis, M. R., & Cullen, J. J. (2002). Assessment of the relationships betweendominant cell size in natural phytoplankton communities and the spectral shape ofthe absorption coefficient. Limnology and Oceanography, 47, 404–417.

Clancy, R. M., & Sadler, W. D. (1992). The fleet numerical oceanography center suite ofoceanographic models and products. Weather and Forecasting, 7, 307–327.

Dall'Olmo, G., Westberry, T. K., Behrenfeld, M. J., Boss, E., & Slade, W. H. (2009).Significant contribution of large particles to optical backscattering in the openocean. Biogeosciences, 6, 947–967.

Dall'Olmo, G., Boss, E., Behrenfeld, M. J., Westberry, T. K., Courties, C., Prieur, L., et al.(2011). Inferring phytoplankton carbon and eco-physiological rates from dielcycles of spectral particulate beam-attenuation coefficient. Biogeosciences, 8,3423–3439.

Dierssen, H. M. (2010). Perspectives on empirical approaches for ocean color remotesensing of chlorophyll in a changing climate. Proceedings of the National Academyof Sciences of the United States of America, 107, 17073–17078.

Doney, S. C. (2006). Oceanography: Plankton in a warmer world. Nature, 444, 695–696.DuRand, M. D., & Olson, R. J. (1996). Contributions of phytoplankton light scattering

and cell concentration changes to diel variations in beamattenuation in the equatorialPacific from flow cytometric measurements of pico-, ultra-, and nanoplankton.Deep-Sea Research Part II, 43, 891–906.

Eplee, R. E., Jr., Meister, G., Patt, F. S., Franz, B. A., & McClain, C. R. (2011). Uncertaintyassessment of the SeaWiFS on-orbit calibration, in Earth Observing Systems XVI. InJ. J. Butler, X. Xiong, & X. Gu (Eds.), Proc. SPIE, 8153. (pp. 815310).

Falkowski, P. G. (1984). Physiological responses of phytoplankton to natural lightregimes. Journal of Plankton Research, 6, 295–307.

Falkowski, P. G., Barber, R. T., & Smetacek, V. (1998). Biogeochemical controls and feed-backs on ocean primary production. Science, 281, 200–206.

Field, C. B., Behrenfeld, M. J., Randerson, J. T., & Falkowski, P. G. (1998). Primary productionof the biosphere: Integrating terrestrial and oceanic components. Science, 281,237–240.

Franz, B. A., Bailey, S. W., Werdell, P. J., & McClain, C. R. (2007). Sensor-independent ap-proach to vicarious calibration of satellite ocean color radiometry. Applied Optics,46, 5068–5082.

Gernez, P., Antoine, D., & Huot, Y. (2011). Diel cycles of the particulate beam attenuationcoefficient under varying trophic conditions in the northwestern Mediterranean Sea:Observations and modeling. Limnology and Oceanography, 56, 17–36.

Gordon, H. R. (1997). Atmospheric correction of ocean color imagery in the EarthObserving System era. Journal of Geophysical Research, 102, 17081–17106.

Graff, J. R., Milligan, A. J., & Behrenfeld, M. J. (2012). The measurement of phytoplanktonbiomass using flow-cytometric sorting and elemental analysis of carbon. Limnologyand Oceanography: Methods, 10, 910–920.

Green, R. E., Sosik, H. M., Olson, R. J., & DuRand,M. D. (2003). Flow cytometric determinationof size and complex refractive index for marine particles: Comparison with indepen-dent and bulk estimates. Applied Optics, 42, 526–541.

Gregg, W. W., Casey, N. W., & McClain, C. R. (2005). Recent trends in global ocean chloro-phyll. Geophysical Research Letters, 32 (L03606).

Gregg, W. W., Conkright, M. E., Ginoux, P., O'Reilly, J. E., & Casey, N. W. (2003). Oceanprimary production and climate: Global decadal changes. Geophysical ResearchLetters, 30, 1809. http://dx.doi.org/10.1029/2003GL016889.

Halsey, K. H., Milligan, A., & Behrenfeld, M. J. (2010). Physiological optimization under-lies growth rate-independent chlorophyll-specific gross and net primary produc-tion. Photosynthesis Research, 103, 125–137. http://dx.doi.org/10.1007/s11120-009-9526-z.

Henson, S. A., Sarmiento, J. L., Dunne, J. P., Bopp, L., Lima, I., Doney, S. C., et al. (2010).Detection of anthropogenic climate change in satellite records of ocean chlorophylland productivity. Biogeosciences, 7, 621–640.

Higgins, H. W., Wright, S. W., & Schülter, L. (2011). Quantitative interpretation of che-motaxonomic pigment data. In S. Roy, C. A. Llewellyn, E. S. Egeland, & G. Johnsen(Eds.), Phytoplankton pigments (pp. 257–313). : SCOR — Cambridge UniversityPress.

Hu, C., Carder, K. L., & Muller-Karger, F. E. (2001). How precise are SeaWiFS ocean colorestimates? Implications of digitization-noise errors. Remote Sensing of Environment,76, 239–249.

Huot, Y., Morel, A., Twardowski, M. S., Stramski, D., & Reynolds, R. A. (2008). Particleoptical backscattering along a chlorophyll gradient in the upper layer of the easternSouth Pacific Ocean. Biogeosciences, 5, 495–507.

IOCCG (2006). In ZhongPing Lee (Ed.), Remote sensing of inherent optical properties:Fundamentals, tests of algorithms, and applications. Report #5 (pp. 126)(http://www.ioccg.org/reports/report5.pdf)

Jeong, M. -J., Hsu, N. C., Kwiatkowska, E. J., Franz, B. A., Meister, G., & Salustro, C. E.(2011). Impacts of cross-platform vicarious calibration on the deep blue aerosolretrievals formoderate resolution imaging spectroradiometer aboard Terra. Transactionson Geoscience and Remote Sensing, 40(12).

Johnson, K. S., Berelson, W. M., Boss, E. S., Chase, Z., Claustre, H., Emerson, S. R., et al.(2009). Observing biogeochemical cycles at global scales with profiling floats andgliders: Prospects for a global array. Oceanography, 22, 216–225.

Kirk, J. T. O. (1994). Light and photosynthesis in aquatic ecosystems (2nd ed.). Cambridge,U.K.: Cambridge Univ. Press (509 pp).

Kitidis, V., Stubbins, A., Uher, G., Upstill-Goddard, R., Law, C., & Woodward, E. (2006).Variability of chromophoric organic matter in surface waters of the AtlanticOcean. Deep-Sea Research Part II, 53, 1666–1684.

Kostadinov, T. S., Siegel, D. A., & Maritorena, S. (2010). Global variability of phyto-plankton functional types from space: Assessment via the particle size distribu-tion. Biogeosciences, 7, 3239–3257.

Kwiatkowska, E. J., Franz, B. A., Meister, G., McClain, C., & Xiong, X. (2008). Cross-calibrationof ocean-color bands from Moderate Resolution Imaging Spectroradiometer on Terraplatform. Applied Optics, 47(36).

Laws, E. A., & Bannister, T. T. (1980). Nutrient- and light-limited growth of Thalassiosirafluviatilis in continuous culture, with implications for phytoplankton growth in theocean. Limnology and Oceanography, 25, 457–473.

Lee, Z. P., Carder, K. L., & Arnone, R. (2002). Deriving inherent optical properties fromwater color: A multi-band quasi-analytical algorithm for optically deep waters.Applied Optics, 41, 5755–5772.

Lee, Z., Shang, S., Hu, C., Lewis, M., Arnone, R., Li, Y., et al. (2010). Time series ofbio-optical properties in a subtropical gyre: Implications for the evaluation ofinterannual trends of biogeochemical properties. Journal of Geophysical Research,115. http://dx.doi.org/10.1029/2009JC005865 (C09012).

Maritorena, S., Fanton d'Andon, O. H., Mangin, A., & Siegel, D. A. (2010). Merged satelliteocean color data products using a bio-optical model: Characteristics, benefits andissues. Remote Sensing of Environment, 114, 1791–1804.

Maritorena, S., Siegel, D. A., & Peterson, A. R. (2002). Optimization of a semianalyticalocean color model for global-scale applications. Applied Optics, 41, 2705–2714.

Martinez, E., Antoine, D., D'Ortenzio, F., & Gentili, B. (2009). Climate-driven basin-scaledecadal oscillations of oceanic phytoplankton. Science, 326, 1253–1256.

McClain, C. R. (2009). A decade of satellite ocean color observations. Annual Review ofMarine Science, 1, 19–42.

McClain, C. R., Feldman, G. C., & Hooker, S. B. (2004). An overview of the SeaWiFSproject and strategies for producing a climate research quality global ocean bio-opticaltime series. Deep-Sea Research Part II, 51, 5–42.

Meister, G., Franz, B. A., Kwiatkowska, E. J., & McClain, C. R. (2012). Corrections to thecalibration ofMODIS aqua ocean color bands derived from SeaWiFS data. Transactionson Geoscience and Remote Sensing, 50(1).

Millet, D. B., Guenther, A., Siegel, D. A., Nelson, N. B., Singh, H. B., de Gouw, J. A., et al. (2010).Global atmospheric budget of acetaldehyde: 3D model analysis and constraintsfrom in-situ and satellite observations. Atmospheric Chemistry and Physics, 10,3405–3425.

Morel, A. (1988). Optical modeling of the upper ocean in relation to its biogenous mattercontent (case 1 water). Journal of Geophysical Research, 93, 10,749–10,768.

Morel, A., & Bricaud, A. (1981). Theoretical results concerning light absorption in adiscrete medium, and application to the specific absorption of phytoplankton. DeepSea Research, 28, 1375–1393.

Morel, A., Claustre, H., & Gentili, B. (2010). The most oligotrophic subtropical zones ofthe global ocean: Similarities and differences in terms of chlorophyll and yellowsubstance. Biogeosciences, 7, 3139–3151.

http://dx.doi.org/10.1029/2004GB002299

http://dx.doi.org/10.1098/rstb.2008.0019

http://dx.doi.org/10.1038/nature09953

http://dx.doi.org/10.1029/2010GB003952

http://dx.doi.org/10.1029/2004JC002419

http://dx.doi.org/10.1029/2004JC002419

http://dx.doi.org/10.1029/2003GL016889

http://dx.doi.org/10.1007/s11120-009-9526-z

http://dx.doi.org/10.1007/s11120-009-9526-z

http://www.ioccg.org/reports/report5.pdf

http://dx.doi.org/10.1029/2009JC005865


Morel, A., Huot, Y., Gentili, B., Werdell, P., Hooker, S., & Franz, B. (2007). Examining the con-sistency of products derived from various ocean color sensors in open ocean (case 1)waters in the perspective of a multi-sensor approach. Remote Sensing of Environment,111, 69–88.

Mouw, C. B., & Yoder, J. A. (2010). Optical determination of phytoplankton sizedistribution from global SeaWiFS imagery. Journal of Geophysical Research, 115.http://dx.doi.org/10.1029/1020JC006337 (C12018).

National Research Council (2011). Assessing the requirements for sustained ocean colorresearch and operations. Washington D.C.: National Academy Press (148 pp).

Nelson, N. B., Carlson, C. A., & Steinberg, D. K. (2004). Production of chromophoricdissolved organic matter by Sargasso Sea microbes.Marine Chemistry, 89, 273–287.

Nelson, N. B., & Siegel, D. A. (2002). Chromophoric DOM in the open ocean. In D. A.Hansell, & C. A. Carlson (Eds.), Biogeochemistry of marine dissolved organic matter(pp. 547–578). San Diego, CA: Academic Press.

Nelson, N. B., & Siegel, D. A. (2013). The global distribution and dynamics of chromophoricdissolved organic matter. Annual Review of Marine Science, 5, 447–476.

Nelson, N. B., Siegel, D. A., Carlson, C. A., & Swan, C. M. (2010). Tracing global biogeochemicalcycles and meridional overturning circulation using chromophoric dissolved organicmatter. Geophysical Research Letters, 37. http://dx.doi.org/10.1029/2009GL042325(L03610).

Nelson, N. B., Siegel, D. A., Carlson, C. A., Swan, C., Smethie, W. M., Jr., & Khatiwala, S.(2007). Hydrography of chromophoric dissolved organic matter in the NorthAtlantic. Deep-Sea Research Part I, 54, 710–731.

Nelson, N. B., Siegel, D. A., & Michaels, A. F. (1998). Seasonal dynamics of coloreddissolved material in the Sargasso Sea. Deep-Sea Research Part I, 45, 931–957.

O'Reilly, J. E., Maritorena, S., Mitchell, B. G., Siegel, D. A., Carder, K. L., Kahru, M., et al.(1998). Ocean color chlorophyll algorithms for SeaWiFS. Journal of GeophysicalResearch, 103, 24,937–24,953.

PACE SDT (). A report of the Pre-Aerosol, Clouds, and ocean Ecosystem (PACE) missionscience definition team. http://decadal.gsfc.nasa.gov/PACE.html (302 pp., Available at)

Polovina, J. J., Howell, E. A., & Abecassis, M. (2008). Ocean's least productive waters areexpanding. Geophysical Research Letters, 35. http://dx.doi.org/10.1029/2007GL031745(L03618).

Rayner, N. A., Parker, D. E., Horton, E. B., Folland, C. K., Alexander, L. V., Rowell, D. P., et al.(2003). Global analyses of sea surface temperature, sea ice, and night marine airtemperature since the late nineteenth century. Journal of Geophysical Research, 108.

Sauer, M. J., Roesler, C. S., Werdell, P. J., & Barnard, A. (2012). Under the hood of satelliteempirical chlorophyll a algorithms: Revealing the dependencies of maximum bandratio algorithms on inherent optical properties. Optics Express, 20, 20920–20933.

Siegel, D. A., Behrenfeld, M. J., Maritorena, S., O'Malley, R. T., & Fields, E. (2011). Globalocean phytoplankton [in “State of the Climate in 2010”]. Bulletin of the AmericanMeteorological Society, 92, S105–S108.

Siegel, D. A., Dickey, T. D., Washburn, L., Hamilton, M. K., & Mitchell, B. G. (1989).Optical determination of particulate abundance and production variations in theoligotrophic ocean. Deep Sea Research, 36, 211–222.

Siegel, D. A., Doney, S. C., & Yoder, J. A. (2002a). The spring bloom of phytoplankton inthe North Atlantic Ocean and Sverdrup's critical depth hypothesis. Science, 296,730–733.

Siegel, D. A., & Franz, B. A. (2010). Oceanography: A century of phytoplankton change.Nature, 466, 569–570.

Siegel, D. A., Maritorena, S., Nelson, N. B., & Behrenfeld, M. J. (2005a). Independence andinterdependencies of global ocean color properties; Reassessing the bio-optical assump-tion. Journal of Geophysical Research, 110. http://dx.doi.org/10.1029/2004JC002527(C07011).

Siegel, D. A., Maritorena, S., Nelson, N. B., Behrenfeld, M. J., & McClain, C. R. (2005b). Coloreddissolved organic matter and its influence on the satellite-based characterizationof the ocean biosphere. Geophysical Research Letters, 32. http://dx.doi.org/10.1029/2005GL024310 L20605.

Siegel, D. A., Maritorena, S., Nelson, N. B., Hansell, D. A., & Lorenzi-Kayser, M. (2002b).Global distribution and dynamics of colored dissolved and detrital organic materials.Journal of Geophysical Research, 107, 3228. http://dx.doi.org/10.1029/2001JC000965.

Smith, R. C., & Baker, K. S. (1978). The bio-optical state of ocean waters and remotesensing. Limnology and Oceanography, 23, 247–259.

Stramski, D., Boss, E., Bogucki, D., & Voss, K. J. (2004). The role of seawater constituentsin light backscattering in the ocean. Progress in Oceanography, 61, 27–56.

Stramski, D., & Kiefer, D. A. (1991). Light scattering by microorganisms in the openocean. Progress in Oceanography, 28, 343–383.

Sverdrup, H. U. (1955). The place of physical oceanography in oceanographic research.Journal of Marine Research, 14, 287–294.

Swan, C. M., Nelson, N. B., Siegel, D. A., & Kostadinov, T. S. (2012). The effect of surfaceirradiance on the absorption spectrum of chromophoric dissolved organic matterin the global ocean. Deep-Sea Research Part I, 63, 52–64.

Swan, C. M., Siegel, D. A., Nelson, N. B., Carlson, C. A., & Nasir, E. (2009). Biogeochemicaland hydrographic controls on chromophoric dissolved organic matter distributionin the Pacific Ocean. Deep-Sea Research Part I, 56, 2172–2192.

Szeto, M., Werdell, P. J., Moore, T. S., & Campbell, J. W. (2011). Are the world's oceansoptically different? Journal of Geophysical Research, 116. http://dx.doi.org/10.1029/2011JC007230.

Toole, D. A., Siegel, D. A., & Doney, S. C. (2008). A light-driven, one-dimensionaldimethylsulfide biogeochemical cycling model for the Sargasso Sea. Journal ofGeophysical Research, 113. http://dx.doi.org/10.1029/2007JG000426 (G02009).

Torrecilla, E., Stramski, D., Reynolds, R. A., Millán-Núñez, E., & Piera, J. (2011). Clusteranalysis of hyperspectral optical data for discriminating phytoplankton pigmentassemblages in the open ocean. Remote Sensing of Environment, 115, 2578–2593.

Vantrepotte, V., & Mélin, F. (2009). Temporal variability of 10-year global SeaWiFStime-series of phytoplankton chlorophyll a concentration. ICES Journal of MarineScience: Journal du Conseil. http://dx.doi.org/10.1093/icesjms/fsp107.

Vantrepotte, V., &Mélin, F. (2011). Inter-annual variations in the SeaWiFS global chlorophylla concentration (1997–2007). Deep-Sea Research Part I, 58, 429–441.

Werdell, P. J., & Bailey, S. W. (2005). An improved bio-optical data set for ocean coloralgorithm development and satellite data product validation. Remote Sensing ofEnvironment, 98, 122–140.

Westberry, T. K., Behrenfeld, M. J., Siegel, D. A., & Boss, E. (2008). Carbon-based primaryproductivitymodelingwith vertically resolved photoacclimation.Global BiogeochemicalCycles, 22. http://dx.doi.org/10.1029/2007GB003078 (GB2024).

Westberry, T. K., Dall'Olmo, G., Boss, E., Behrenfeld, M. J., & Moutin, T. (2010). Coherenceof particulate beam attenuation and backscattering coefficients in diverse open oceanenvironments. Optics Express, 18, 15419–15425.

Xue, Y., Reynolds, R. W., Banzon, V., Smith, T. M., & Rayner, N. A. (2011). Sea surfacetemperatures in 2010. In “State of Climate 2010”. Bulletin of the AmericanMeteorologicalSociety, 92, S78–S81.

Yamashita, Y., & Tanoue, E. (2009). Basin scale distribution of chromophoric dissolvedorganic matter in the Pacific Ocean. Limnology and Oceanography, 54, 598–609.

Yoder, J. A., & Kennelly, M. A. (2003). Seasonal and ENSO variability in global oceanphytoplankton chlorophyll derived from 4 years of SeaWiFS measurements. GlobalBiogeochemical Cycles, 17. http://dx.doi.org/10.1029/2002GB001942.

Yoder, J., McClain, C. R., Feldman, G. C., & Esaias, W. E. (1993). Annual cycles of phyto-plankton chlorophyll concentrations in the global ocean: A satellite view. GlobalBiogeochemical Cycles, 7, 181–193.

Zepp, R. G., Erickson, D. J., III, Paul, N. D., & Sulzberger, B. (2007). Interactive effects ofsolar UV radiation and climate change on biogeochemical cycling. Photochemical& Photobiological Sciences, 6, 286–300.

http://dx.doi.org/10.1029/1020JC006337

http://dx.doi.org/10.1029/2009GL042325

http://decadal.gsfc.nasa.gov/PACE.html

http://dx.doi.org/10.1029/2007GL031745

http://dx.doi.org/10.1029/2004JC002527

http://dx.doi.org/10.1029/2005GL024310

http://dx.doi.org/10.1029/2005GL024310

http://dx.doi.org/10.1029/2001JC000965

http://dx.doi.org/10.1029/2011JC007230

http://dx.doi.org/10.1029/2011JC007230

http://dx.doi.org/10.1029/2007JG000426

http://dx.doi.org/10.1093/icesjms/fsp107

http://dx.doi.org/10.1029/2007GB003078

http://dx.doi.org/10.1029/2002GB001942

Remote Sensing of Environment - misclab.umeoce.maine.edumisclab.umeoce.maine.edu/documents/Siegeletal2003.pdf · from satellite ocean color sensors that sample the entire Earth surface

Documents